<kof673>
nikola: i was going to use crit bit tree for a long time for string data...but i had some named things, that need to lookup by name (string) and also see if name is already in use...better than blindly iterating e.g. an array at least. not a problem now, but 10, 100, 1000 entries...
<kof673>
even the simplest binary tree is better than that, does not have to potentially look at all entries
<nikolapdp>
yeah that's fair
<nikolapdp>
i'd probably make a basic hash map in that case
Vercas3 has quit [Remote host closed the connection]
Vercas3 has joined #osdev
gbowne1 has quit [Remote host closed the connection]
gbowne1 has joined #osdev
Arthuria has quit [Ping timeout: 268 seconds]
netbsduser has joined #osdev
rustyy has quit [Ping timeout: 240 seconds]
rustyy has joined #osdev
netbsduser has quit [Ping timeout: 246 seconds]
randm has quit [Ping timeout: 255 seconds]
sjs has quit [Quit: sjs]
tjf has quit [Ping timeout: 268 seconds]
sjs has joined #osdev
randm has joined #osdev
tjf has joined #osdev
duckworld has quit [Ping timeout: 250 seconds]
duckworld has joined #osdev
tbvdm has joined #osdev
bliminse has quit [Quit: leaving]
goliath has joined #osdev
aejsmith has joined #osdev
nikolapdp has quit [Ping timeout: 246 seconds]
bliminse has joined #osdev
TkTech9 has joined #osdev
TkTech has quit [Ping timeout: 256 seconds]
TkTech9 is now known as TkTech
janemba has quit [Read error: Connection reset by peer]
gbowne1 has quit [Remote host closed the connection]
Left_Turn has joined #osdev
Turn_Left has joined #osdev
Left_Turn has quit [Ping timeout: 268 seconds]
obrien has joined #osdev
GeDaMo has joined #osdev
osdev199 has joined #osdev
<osdev199>
Hi, how does an NVMe controller read from or write to the queues? Can a kernel using UEFI page table restrict the read or write of the controller to an allocated memory area for queues?
nikolapdp has joined #osdev
nikolapdp has quit [Remote host closed the connection]
nikolapdp has joined #osdev
<Mutabah>
The controller uses bus-mastering DMA to read/write directly to RAM
<Mutabah>
This uses physical addressing, and thus ignores the CPU's page tables
<Mutabah>
however, there's a feature called the "IOMMU" that allows you to apply paging to a specific device
<nikolapdp>
what do you mean by "apply paging"
osdev199 has quit [Remote host closed the connection]
gsekulski has joined #osdev
osdev199 has joined #osdev
<Mutabah>
nikolapdp: Map logical addresses (that the device sees/requests) into a different physical address (on RAM)
<Mutabah>
same as paging does for the CPU's view of memory
<nikolapdp>
so basically move it elsewhere in the physical address space
<Mutabah>
Yup
<Mutabah>
Actually... depends on what you mean by that
<Mutabah>
it doesn't move where the device's registers are
<Mutabah>
it prevents an untrusted device from accessing all of RAM
<nikolapdp>
yeah i've heard of iommu but i don't know what exactly it does
<Ermine>
So basically iommu enables devices to use virtual addresses?
<heat>
solaris internals the great venerable sun microsystems did this thing but now we figured out it was bad and now we have this innovative mega cool system
<heat>
praise the sun
<nikolapdp>
heat what did he say about ffs
<heat>
i dont know i didn't read the 4.4BSD nor FreeBSD books
<heat>
it was probably the bees' knees
<nikolapdp>
kek
<heat>
mjg, did you know that system v brought us hash tables
<mjg>
nice invention
<kof673>
gkrellm had a flynn plugin aka doom guy
<heat>
yeah
<heat>
and buffer and inode caches
<mjg>
did you know that literally nobody allocated objects of the same size one after another
<mjg>
before bonwick
<heat>
release 2 brought us demand paging and cow
<heat>
approaching optimal!
<mjg>
careful tho, the G man may start picking on you for shitting oon solaris
<heat>
garry's mod doesn't run on solaris i think
<mjg>
oh i just remember a proper 00s linux screen shot, apart from gkrellm, must have had tail -f /var/log/messages in a transparent terminal
<vdamewood>
Oracle shit on Solaris enough for everyone.
<mjg>
bonus points if the wallpaper made things hard to read
<heat>
i was wondering how usable svr4 is in 2024, with 2024 programs
<heat>
i quickly realized it doesn't have threads
<heat>
so there's probably the big problem right there
<mjg>
windows 95 mofo
<mjg>
one day i'll will be arsed enough to get a working compiler there and bench reading files 'n shit
<mjg>
vs era-appropriate linux
<mjg>
no apropriate hw though 8(
<zid>
for fuck sake?
<nikolar>
I'm curious how those would compare
<mjg>
i expect both to suck, linux always had some of the stuff done wrong in the name of optimization
<mjg>
i mean they unknowingly slowed it down
<nikolar>
Also throw in a Unix like Solaris into that too
q3lont has joined #osdev
navi has joined #osdev
m3a has quit [Ping timeout: 268 seconds]
zxrom has joined #osdev
q3lont has quit [Quit: Leaving]
junon has joined #osdev
junon has quit [Remote host closed the connection]
Arthuria has joined #osdev
nur has quit [Read error: Connection reset by peer]
Arthuria has quit [Killed (NickServ (GHOST command used by Guest684531))]
Arthuria has joined #osdev
netbsduser has joined #osdev
m257 has joined #osdev
Arthuria has quit [Ping timeout: 240 seconds]
eddof13 has joined #osdev
eddof13 has quit [Client Quit]
m257 has quit [Quit: Client closed]
m257 has joined #osdev
FreeFull has quit []
vdamewood has quit [Quit: My Mac has gone to sleep. ZZZzzz…]
bauen1 has quit [Ping timeout: 246 seconds]
FreeFull has joined #osdev
obrien has quit [Remote host closed the connection]
spareproject has joined #osdev
eddof13 has joined #osdev
eddof13 has quit [Client Quit]
adder has quit [Remote host closed the connection]
<Ermine>
Also intel did holographic stickers and they look cool
<heat>
yeah they look nice
<Ermine>
looked*
<heat>
i didn't update my arch linux for some weeks and now i have 533 packages to update
<nikolapdp>
those are rookie numbers
<nikolapdp>
only 500 after a few weeks
eddof13 has quit [Client Quit]
* geist
yawns
<nikolapdp>
well herllo there geist
<Ermine>
if you don't update it for months you'd need to upgrade keyring manually
<mjg>
arch linux any gud?
<mjg>
genuine question
<mjg>
no memes
<gog>
it's linux
<Ermine>
If not my laziness, I'd switch from my opensuse to arch
<nikolapdp>
arch is pretty good
<mjg>
for example, do you texpet stuff to break any time you update
<nikolapdp>
no
<Ermine>
my genuine optinion
<nikolapdp>
it very rarely happens,
<nikolapdp>
like exceedingly rare
<Ermine>
arch is pretty stable actually
<nikolapdp>
yeah
<gog>
they announce when an update is expected to cause breakage but i've almost never experienced it
<heat>
^^and that mofo is using fucking artix
<Ermine>
(except aur stuff)
<heat>
^^and that mofo is using fucking manjaro
<nikolapdp>
see
<gog>
i'm not using manjaro
<gog>
i'm on real archj
<heat>
so imagine how stable the arch linux experience is!
<nikolapdp>
arch family is rock solid
<heat>
oh you switched?
<Ermine>
i don't use artix
<gog>
yes i've been on mainline arch for a few months now
<heat>
congrats
<nikolapdp>
not you Ermine
<gog>
i just deleted my manjaro partition
<heat>
goated
<nikolapdp>
good
<gog>
i finally worked out my boot issues
<gog>
i forgot to add -U to genfstab
<Ermine>
manjaro is shitty, on the other hands. It's on my blacklist
<nikolapdp>
yeah it sucks
<heat>
i run alpine on my rpi
<gog>
manjaro worked fine for me tbh
<nikolapdp>
a lot of issues are self inflicted on htheir side
<Ermine>
other hand*
<heat>
MUSL!
<gog>
i never had issues with it
<Ermine>
better run onyx on it
<gog>
but i was a little tired of waiting for updates
<Ermine>
btw how much do rpis cost for you?
<heat>
Ermine, i would do that but i dont have wifi support and arm64 is not quite stable
<Ermine>
ah, wifi
<heat>
yep
<gog>
anyhow, linux is linux and if you use linux you're going to have a bad time in some way
<gog>
choose your poison
<mjg>
8(
* mjg
sticks to his solaris 7 on sparcstation
<gog>
i guess one could say the same about any operating system
<heat>
uhhh a rpi 4b 2G is 50 euro here
<gog>
except illumos
<heat>
my rpi zero 2 w was like 15 euro, pretty cheap
<nikolapdp>
ILLUMOS
<heat>
gog, except ONYX
<mjg>
how do you know onyx neer breaks
<mjg>
it already does not work
<heat>
im the onyx user
<mjg>
:rimshot:
<heat>
oh an rpi 5 4G is 66 euro
<Ermine>
heat: that model costs twice more here
<Ermine>
that = 4b 2g
janemba has joined #osdev
<heat>
the gang gets sanctioned
<Ermine>
+ parallel import
<heat>
i love the european union i love the single market
<heat>
schengen area wooooooo
<mjg>
easterner of the west
<nikolapdp>
kek it's 92 euro here
<Ermine>
mjg: how much does it cost for you?
<mjg>
fuck if i know
<mjg>
it would be counted in weed grams anyway
<heat>
Ermine, looks roughly the same as my prices
<heat>
but in polish coin so x5
<heat>
SINGLE MARKET BABY WOOOOO EUROPEAN UNION SCHENGEN AREA LETS GOOOOOOOOO
<Ermine>
I had schengen visa once
<gorgonical>
how we doin today kernal developers
<gorgonical>
I've had this thought floating around in my head for a while that I can't decide the value of
<nikolapdp>
hello there gorgonical
<gorgonical>
the basic gist: when you make a syscall, that's a request to the kernel for some kind of privileged service. I wondered briefly, let's imagine you have many kernels on a system. Is there any reason you might want to change which kernel services a syscall?
<gorgonical>
I think this idea is sort of approximated by userland servers in microkernels, but I don't think the matchup is exactly the same
<Ermine>
hi gorgonical
<heat>
>let's imagine you have many kernels on a system
<heat>
no.
<heat>
hth
<gorgonical>
this is a common occurrence for me
<Ermine>
exokernals go!
<heat>
one kernel is a kernel too much
<nikolapdp>
just poke the hardware directly
<gorgonical>
be your own kernel
<heat>
do not
<heat>
rock and fire
<Ermine>
Otoh first type hypervisors do exist
<geist>
type 1 yeah
<geist>
they're not as in vogue anymore i think
<gorgonical>
Not so much, no
<Ermine>
though I only know of ESXi and Xen
<gorgonical>
That was my first publication, though
<Ermine>
But really, your idea looks just like an exokernel
<gorgonical>
In what way? I thought exokrenels were sort of per-hardware device microkernels
<gorgonical>
Basically just implementing concurrency locks and such
<Ermine>
afaik they delegate most of the stuff to libOSen in the userspace
<Ermine>
So your idea can be implementing as something sitting in an exokernel and dispatching syscalls to some libOS
<Ermine>
but anyway, this seems way to theoretical, and I don't know how it can be useful
<gorgonical>
I don't either
<gorgonical>
I'm puzzling about it. I can think of maybe a few situations, though. Kernels implementing their own policy on hardware means you could have kernels in charge of DRAM banks. One kernel does demand paging, one does contiguous pinning. An mmap can be directed depending on the needs of the application.
<alice>
gorgonical, You might possibly find the MS research paper on Drawbridge interesting there (or, at least, the 'picoprocess' part of them). I believe that gets implemented as a small shim that captures 'syscall' and tunnels to a different subsystem to actually handle it. -- https://www.microsoft.com/en-us/research/wp-content/uploads/2016/02/asplos2011-drawbridge.pdf
<gorgonical>
alice: thanks, I'll read this. I've heard of this paper but never read it. I don't have much exposure to library OS research
<netbsduser>
GeDaMo: finally these things are starting to get to semi-reasonable prices
<netbsduser>
too bad it will probably be a pile of proprietary junk
<nikolapdp>
what are you talking about netbsdusers
<GeDaMo>
Probably those RISC-V laptops
<nikolapdp>
oh yea
<zid>
GeDaMo: What obscure puzzle game today?
<GeDaMo>
Sorry, just Quordle and Quordle sequence :P
<zid>
that definitely isn't a wordle word normally
<zid>
they must have their own wordlist
<GeDaMo>
Given that it's a dictionary company, that wouldn't be a surprise :P
<geist>
yah re: that syscall trapping thing we have something in fuchsia called Restricted Mode (i have a patent on it even)
<geist>
lets us run linux binaries directly on zircon kernel
node1 has joined #osdev
<gorgonical>
so restricted mode creates a kernel-userspace split inside of an existing zircon process?
<gorgonical>
nesting the distinction?
<gorgonical>
so that the starnix kernel that acts as a linux "personality" can run safely inside the zircon process to handle the linux syscalls and such
<nikolar>
geist: ooh can do i have permission to use that
<nikolar>
s/can//
<geist>
gorgonical: basically. and the trick is you can map the same starnix kernel in multiple address spaces
<geist>
ie, the top half of the user aspace is sharing the same page tables
<gorgonical>
yeah indeed, just the same as maybe a vdso or the kernel pagetables in the upper half
<gorgonical>
of a regular linux process
Arthuria has joined #osdev
<geist>
yah the only limit is the linux process sees something that's half the size
<gorgonical>
sure
<geist>
ie, one bit is chewed up with the starnix half
<gorgonical>
the patent abstract is broad enough to cover anything you could characterize as a hypervisor imo
<nikolapdp>
so geist, as a patent holder, do i have permission to use it in my projects :P
<gorgonical>
he's not the holder, google is
<geist>
heh though really google is pretty chill about it
<geist>
actually i dunno if the patent went through
<geist>
it's probabyl still in progress
<nikolapdp>
oh you said "i have" so i though it was your name on it lol
<gorgonical>
it is
<gorgonical>
as inventor
<nikolapdp>
but not holder
<gorgonical>
I don't know how to read patents so I can't tell if this is finished or in progress or what
<zid>
You can enforce it in googlecourt
<geist>
onestly i'm not really proud of the patents, i'm not a big fan of software patents
<zid>
your premier litigation solution
<geist>
but sometimes company does stuff so there ya go
<zid>
please drink the verification can to proceed with the lawsuit
<gorgonical>
i applied for a job earlier this year as a research scientist and they said pretty directly that their measure of your productivity was the product of your research publications and patents filed
<gorgonical>
with a greater bias towards patents
<zid>
Yea academia has been like that for a good while now
<zid>
hence the current replication crisis
<gorgonical>
this was a company research division actually
<zid>
nobody publishes negative results, nobody reproduces results, no corporate money in it, so no jobs in it
<zid>
almost all papers are just total bollocks nobody has ever checked, because without it someone's metrics would be worse and they'd lose their funding and/or job
<nikolar>
Yeah
<nikolar>
It's great isn't it
<gorgonical>
geist: im curious, how did they decide that this was patentable and distinct enough from existing hypervisor-type compatibilty approaches?
<gorgonical>
like, did they press you and ask if you thought it was patentable? or did someone else see the work and get the idea?
<zid>
patent first, litigate whether it's valid later
<geist>
gorgonical: kinda. though i probably shouldn't talk about it
<geist>
mostly lawyers did the work
<gorgonical>
i understand
nikolapdp has quit [Ping timeout: 264 seconds]
<gorgonical>
the job I mentioned made it clear that we were responsible for "patent seeking" in a way, framing work as patentable units of IP to protect
<gorgonical>
In a similar vein to how academics are incentivized to view work as "minimum publishable units" instead of a grander arc of research
<mjg>
"a retangular communication device"
<GeDaMo>
With rounded corners! :P
nikolapdp has joined #osdev
<nikolapdp>
sadge
<Bitweasil>
On the internet! :p
eddof13 has joined #osdev
spareproject has quit [Remote host closed the connection]
node1 has quit [Quit: Client closed]
xenos1984 has quit [Read error: Connection reset by peer]
gsekulski has quit [Ping timeout: 256 seconds]
janemba has quit [Read error: Connection reset by peer]
xenos1984 has joined #osdev
gsekulski has joined #osdev
janemba has joined #osdev
gsekulski has quit [Ping timeout: 268 seconds]
mxs9 has joined #osdev
gbowne1 has joined #osdev
gbowne1 has quit [Remote host closed the connection]
gbowne1 has joined #osdev
GeDaMo has quit [Quit: 0wt 0f v0w3ls.]
<heat>
patenten
<nikolapdp>
PATCHEN
<mjg>
don't tell G man, i booted OMNIOS (an ILLUMOS variant) and ran mdb to check some stuff
<mjg>
it segfaulted
<heat>
im a big fan of patents that are GPL only
<heat>
just to fuck with BSD people
<mjg>
:]
<heat>
what's mdb?
<mjg>
there was a story of a bsd-originated nic driver ported to linux
<mjg>
... and relicensed to gpl
<nikolapdp>
unsurprisingly
<mjg>
heat: their kernel (and other stuff) debugger
<mjg>
genuinely nice tool
<heat>
not a crash(1) fan i see
<mjg>
which crash tho mofer
<mjg>
there was old-school crash debugger which was utter garbage
<heat>
all of them
<mjg>
there is also a linux debugger which took the name and which is ok
<heat>
they're all kinda shit i think?
<nikolapdp>
adb?
<heat>
drgn is nice i think
<mjg>
note i'm talking kernel dumps here
<heat>
i'm aware
<mjg>
for usersapce you probably want something genuinely advanced
<mjg>
with "time travel" 'n shit
<heat>
usersapce lol typo
<heat>
LULKEYBOARD
<mjg>
why are you harping on this
<heat>
i'm kind of stupid
<mjg>
> portugal
<mjg>
carry on
<heat>
yeah goes without saying right
<nikolapdp>
kek
Arthuria has quit [Ping timeout: 268 seconds]
<heat>
there has been some fun little breakage lately cuz they switched some page flags to page type defines or whatever
<heat>
and DWARF doesn't capture defines :))
<heat>
so crash and drgn just died as they couldn't decode page flags in general
<mjg>
:d
<heat>
what's extra fun is that kernel coredumping needs the userspace dumping process to decode struct pages to figure out what needs to be saved
<heat>
it's kind of dumb
<mjg>
it is not given that this is done with kexec
<heat>
not necessarily
<mjg>
?
<heat>
you can get a dump with /proc/vmcore
<heat>
sorry, kcore
<heat>
i think the vmcore is for the kexec crash kernel?
<mjg>
if the system is perfectly fine running then there is even less reason to add logic to the kernel tiself
<heat>
you don't know if it is though
<heat>
could've OOPS'ed
<mjg>
you are talking some fucked setup here
<heat>
why
<mjg>
if you want to dump core, the real way is to kexec into the restricted environment
<heat>
it's perfectly normal to not panic and still want a core dump
<heat>
and getting a coredump via cat /proc/kcore > kcore gets you like 128TB of dump
<nikolar>
lol
<heat>
because they don't even attempt to filter out vmalloc space at all, etc
Arthuria has joined #osdev
<mjg>
then don't cat /proc/kcore
<heat>
that's what im saying it's kind of weird
<heat>
you need a special program to decode struct pages and internal data structures and shit, using DWARF
<heat>
yeah they use makedumpfile
Arthuria has quit [Killed (NickServ (GHOST command used by Guest684531))]
Arthuria has joined #osdev
<mjg>
in general if you want a clear coredump and you are worried about completion because the kernel has oopsed
<mjg>
that's even more reasons to kexec
Arthuria has quit [Read error: Connection reset by peer]
<mjg>
let me interrupt with a message from our sponsor